WATER 4.0 – Predictive Framework

Predictive framework for proactive monitoring and maintenance of water distribution networks, developed using LSTM models and distributed evolutionary optimization Island-CPSO.

BattLeDIM L-Town Network
LSTM Architecture
Island-CPSO Optimization

Project Information

  • Category: Artificial Intelligence
  • Role: Research Fellow
  • Institution: University of Enna “Kore”
  • Year: 2025
  • Project: WATER 4.0 – Smart Factory (CUP: B79J24000580005)

Project Description

Objective: Development and validation of a predictive framework for continuous estimation of water losses in distribution networks, aimed at proactive infrastructure monitoring and maintenance.

Dataset: Usage of the BattLeDIM 2020 dataset on the simulated “L-Town” network; multivariate time series (demand, flows, levels, pressures) sampled every 5 minutes.

Methodology: Design of LSTM architectures for time-series regression, automatic hyperparameter optimization via Island-CPSO, a distributed variant of Continuous Particle Swarm Optimization with island paradigm and asynchronous migration.

Implementation: Experimental pipeline for training and validation, time-window management, normalization with StandardScaler, parallelization on multi-core architectures.

Results: Good model generalization, reduced tuning time thanks to Island-CPSO, and improved regression metrics compared to baselines.

Skills Developed: Sequential modeling (LSTM), distributed evolutionary optimization, water dataset analysis, experimentation, and result visualization.